Re: [PATCH v2 2/4] backlight: Expose brightness curve type through sysfs

From: Daniel Thompson
Date: Fri Jun 28 2019 - 04:35:00 EST


On Wed, Jun 26, 2019 at 04:56:11PM +0200, Pavel Machek wrote:
> Hi!
>
> > Export the type of the brightness curve via the new sysfs attribute
> > 'scale'. The value of the attribute may be a simple string like
> > 'linear' or 'non-linear', or a composite string similar to
> > 'compatible' strings of the device tree. A composite string consists
> > of different elements separated by commas, starting with the
> > most-detailed description and ending with the least-detailed one. An
> > example for a composite string is "cie-1931,perceptual,non-linear"
> > This brightness curve was generated with the CIE 1931 algorithm, it
> > is perceptually linear, but not actually linear in terms of the
> > emitted light. If userspace doesn't know about 'cie-1931' or
> > 'perceptual' it should at least be able to interpret the 'non-linear'
> > part.
>
> I'm not sure the comma-separated thing is a good idea. If it is, it should
> go to the Documentation, not to changelog.

So I viewed the comma-separated thing as allow us to describe facts about
the scale used.

In particular I suspect that some controllers will be non-linear *and*
non-perceptual and that some userspaces, particularly those that animate
backlight changes, may care enough about the difference to ask us to add
another fact to the set that describes that scale.

Having said that I do share your concern that the comma-separated list
is overengineered and that all userspaces will end up implementing
something like:

if (strstr("non-linear", scale) {
mode = PERCEPTUAL;
} else if (strstr("unknown", scale) {
mode = use_existing_hueristic();
} else {
mode = LINEAR;
}


> > +What: /sys/class/backlight/<backlight>/scale
> > +Date: June 2019
> > +KernelVersion: 5.4
> > +Contact: Daniel Thompson <daniel.thompson@xxxxxxxxxx>
> > +Description:
> > + Description of the scale of the brightness curve. The
> > + description consists of one or more elements separated by
> > + commas, from the most detailed to the least detailed
> > + description.
> > +
> > + Possible values are:
> > +
> > + unknown
> > + The scale of the brightness curve is unknown.
> > +
> > + linear
> > + The brightness changes linearly in terms of the emitted
> > + light, changes are perceived as non-linear by the human eye.
> > +
> > + non-linear
> > + The brightness changes non-linearly in terms of the emitted
> > + light, changes might be perceived as linear by the human eye.
>
> non-linear is not too useful as described.

Agree.

The idea is that allows a userspace with simple backlight needs to
simple map the brightness property directly to a slider using the
approach above without worrying about perceptual or (possible future)
logarithmic scales. Such an approach won't be perfect but it
probably won't feel horrible for the user either.

Arguably the descriptions should move away from the raw factual
approach and describe what advise the kernel of offering the
userspace.


> > + perceptual,non-linear
> > + The brightness changes non-linearly in terms of the emitted
> > + light, changes should be perceived as linear by the human eye.
> > +
> > + cie-1931,perceptual,non-linear
> > + The brightness curve was calculated with the CIE 1931
> > + algorithm. Brightness changes non-linearly in terms of the
> > + emitted light, changes should be perceived as linear by the
> > + human eye.
>
> Is it useful to know difference between perceptual, and cie-1931?

Depends how assertive the userspaces are!

If they follow the "fix kernel bugs in the kernel" mantra rather than
implement workarounds and heuristics then I suspect it would not be used
much.


> Would it be useful to export absolute values in some well-known units?
>
> If I'm in dark room, I may want 100mW/m^2 of backlight... And it would
> be nice if I could set same backlight intensity on all my devices
> easily.

I'm a little sceptical that we could calibrate an absolute scale on
enough devices for such a property to be useful. I think it would be
"unknown" on almost every system.


Daniel.